Instructions to use gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192")
model = AutoModelForMultimodalLM.from_pretrained("gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192

SGLang

How to use gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192 with Docker Model Runner:
```
docker model run hf.co/gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192
```

Ornith-1.0-397B-FP8-KoreanCoding-Expert192

한국어+코딩 집중 구조적 전문가 가지치기 모델. VLM(이미지 인식) 전문가를 제외하고 192 슬롯을 언어(Korean/Japanese/Chinese) + 코딩에 집중 할당. 한국어 전문가 gate.weight 1.15× 부스트로 라우팅 확률 보정.

핵심 변경 (vs 이전 MultiDomain-Expert192)

항목	이전 (MultiDomain)	이 모델 (KoreanCoding)
한국어 전문가/layer	~24 (가중치 조합)	~52 (경량 할당)
코딩 전문가/layer	~25 (가중치 조합)	100 (경량 할당)
VLM 전문가	포함 (76% 보존)	제외 (언어+코딩에 재할당)
한국어 라우터 부스트	없음	gate.weight × 1.15
할당 방식	가중 결합 (soft)	카테고리별 경량 (hard)

전문가 할당 (레이어당 192)

카테고리	할당	보호 방식
🇰🇷 한국어	top-50 (hard)	gate.weight ×1.15 부스트
🇯🇵 일본어	top-20 (hard)	한국어와 공유 시 부스트
💻 코딩	top-100 (hard)	SWE-bench/HumanEval 보존
🇨🇳 중국어	top-20 (hard)	중국어 생성 보존
🌐 일반	~22 (weight_norm)	영어/기본 추론
🖼️ VLM	제외	Image-to-Text 불가

Usage

python3 -m sglang.launch_server \
    --model-path /path/to/Ornith-1.0-397B-FP8-KoreanCoding-Expert192 \
    --tp 2 --trust-remote-code \
    --reasoning-parser qwen3 --tool-call-parser qwen3_coder \
    --max-model-len 262144 --chunked-prefill-size 8192 \
    --mem-fraction-static 0.88 \
    --host 0.0.0.0 --port 30000

VRAM (2× RTX PRO 6000, 192GB)

구분	용량
가중치 (192 experts FP8)	~152 GB
런타임	~13 GB
KV-Cache	~27 GB (256K × 7 동시)

Spec

512→192 experts (62.5%), 6×32 정렬
Qwen3.5 MoE, Top-10 + 1 Shared, hybrid GDN+GQA
60 layers, 4096 hidden, 262144 context
FP8 compressed-tensors (원본 보존, 재양자화 없음)
VLM 미지원 (이미지 입력 불가, 텍스트 전용)

License

MIT (inherited from Ornith-1.0-397B)

Downloads last month: 18

Safetensors

Model size

155B params

Tensor type

BF16

F8_E4M3

Model tree for gigascake/Ornith-1.0-397B-FP8-KoreanCoding-Expert192

Base model

ornith-ai/Ornith-1.0-397B

Quantized

(19)

this model